This paper introduces a novel self-supervised learning method for completing depth maps of transparent objects, a challenging task for conventional sensors. By simulating transparent object depth deficits within non-transparent regions, the approach significantly reduces reliance on costly labeled data while achieving comparable performance to supervised methods.

This paper explores the marginal benefits of pre-training and intermediate fine-tuning on a small 5M-parameter Vision Transformer for semantic segmentation. It reveals that while pre-training and fine-tuning provide benefits with diminishing returns, inappropriate intermediate fine-tuning can harm downstream performance.